research and statistical methodsarchive.rsna.org/2015/researchandstatisticalmethods.pdfssa11-01...

Research and Statistical Methods

SPGW01A Welcome and Introductory Remarks

SPGW01B Preparing an R01 Research Application

SPGW01C Preparing K Awards

SPGW01D Clinical Trials in Applications

SPGW01E Program Perspectives

SPGW01

NIH Grantsmanship Workshop

Saturday, Nov. 28 1:00PM - 5:00PM Location: E253AB

RS

AMA PRA Category 1 Credits ™: 3.75ARRT Category A+ Credits: 4.00

ParticipantsGayle E. Woloschak, PhD, Chicago, IL (Moderator) Nothing to Disclose

LEARNING OBJECTIVES

1) Gain greater understanding of the NIH grants process: a. understand the process for preparing a research or training grantapplication. b. learn the elements of a competitive grant application. 2) Gain insight into the new features of the NIH reviewprocess. 3) View the review process in action through a mock study section.

Sub-Events

ParticipantsGayle E. Woloschak, PhD, Chicago, IL (Presenter) Nothing to Disclose

LEARNING OBJECTIVES

View learning objectives under main course title.

ParticipantsPratik Mukherjee, MD, PhD, San Francisco, CA (Presenter) Research Grant, General Electric Company; Medical Adivisory Board,General Electric Company;

LEARNING OBJECTIVES


ParticipantsRuth C. Carlos, MD, MS, Ann Arbor, MI (Presenter) Nothing to Disclose

LEARNING OBJECTIVES


Honored Educators

Presenters or authors on this event have been recognized as RSNA Honored Educators for participating in multiple qualifyingeducational activities. Honored Educators are invested in furthering the profession of radiology by delivering high-qualityeducational content in their field of study. Learn how you can become an honored educator by visiting the website at:https://www.rsna.org/Honored-Educator-Award/

Ruth C. Carlos, MD, MS - 2015 Honored Educator

ParticipantsMichael W. Vannier, MD, Chicago, IL (Presenter) Nothing to Disclose

LEARNING OBJECTIVES


Active Handout:Michael Walter Vannier

http://abstract.rsna.org/uploads/2015/15004306/SPGW01D.pdf

Honored Educators


Michael W. Vannier, MD - 2015 Honored Educator

SPGW01F The Process of Review

SPGW01G Questions to the Faculty

SPGW01H Summary and Evaluation Form

ParticipantsAntonio Sastre, PhD, Bethesda, MD (Presenter) Nothing to Disclose

LEARNING OBJECTIVES



LEARNING OBJECTIVES



LEARNING OBJECTIVES



LEARNING OBJECTIVES


SSA11-01 Informatics Keynote Speaker: Role of NLP and Machine Learning in Radiology

Sunday, Nov. 29 10:45AM - 10:55AM Location: S403A

SSA11-02 Virtual Reality Training Modules Hold Potential for IR Education


SSA11-03 Teaching File 2.0: The Next Generation of Radiology Education with Open Web Technologies


SSA11

ISP: Informatics (Education and Research)

Sunday, Nov. 29 10:45AM - 12:15PM Location: S403A

ED RS

AMA PRA Category 1 Credits ™: 1.50ARRT Category A+ Credit: 1.00

ParticipantsGeorge L. Shih, MD, MS, New York, NY (Moderator) Consultant, Image Safely, Inc; Stockholder, Image Safely, Inc; Consultant,Angular Health, Inc; Stockholder, Angular Health, Inc; Peter R. Bream JR, MD, Nashville, TN (Moderator) Consultant, Coeur, Inc; Advisory Panel, CryoLife, IncLuciano M. Prevedello, MD,MPH, Columbus, OH (Moderator) Nothing to Disclose

Sub-Events

ParticipantsGeorge L. Shih, MD, MS, New York, NY (Presenter) Consultant, Image Safely, Inc; Stockholder, Image Safely, Inc; Consultant,Angular Health, Inc; Stockholder, Angular Health, Inc;

AwardsTrainee Research Prize - Fellow

ParticipantsColin J. McCarthy, MD, Boston, MA (Presenter) Nothing to DiscloseAlvin Y. Yu, MD, Boston, MA (Abstract Co-Author) Nothing to DiscloseStephen R. Lee, MD, Boston, MA (Abstract Co-Author) Nothing to DiscloseSynho Do, PhD, Boston, MA (Abstract Co-Author) Nothing to DiscloseSteven L. Dawson, MD, Cambridge, MA (Abstract Co-Author) Consultant, Cook Group Incorporated Royalties, CAE Healthcare, IncRaul N. Uppot, MD, Boston, MA (Abstract Co-Author) Nothing to Disclose

Background

The challenge for all types of interventional radiology (IR) training is hands-on experience. Currently, IR trainees learn proceduresby watching and assisting in cases. We set out to develop and evaluate an immersive virtual reality (VR) simulator, the aim ofwhich was to provide trainees with an introduction to IR procedures. These modules differ from traditional textbooks or onlinelearning sites, since they immerse the participant in the center of activity. Advances in the VR field have allowed for not onlyrealistic display of stereoscopic and 360-degree images, but also the recording of such immersive content.

Evaluation

An introductory IR tutorial was recorded using a dual camera system (GoPro, Inc.) and post-processed to generate a stereoscopic3D tutorial. The content was displayed using a head-mounted VR headset, featuring low-latency gyroscope paired with two lowpersistence OLED displays located immediately in front of the viewer's eyes, blocking out all external visual stimuli (Oculus VR, LLC).Additional tutorials were also constructed from a 7 camera rig (360Heros, Inc.) that captured 360-degree environmental images,allowing full immersion in the IR suite. Footage from 7 individual cameras was fused and synchronized using software (AutopanoVideo, Kolor SARL). Participants were asked to complete a survey after the tutorial, with questions designed around a Likert scale.

Discussion

12 participants agreed to partake in the study, 6 Attending Radiologists, 5 Fellows and 1 Resident. 75% (n = 9) of those surveyedfelt that the immersive VR training module was "good" or "excellent". All participants felt that the VR headset had potential toimprove IR training in the future. Feedback included some reports of initial motion sickness, potential to interact with the 3Denvironment and the need to incorporate radiology images into the interactive content.

Conclusion

Immersive VR platforms hold potential to enhance education in Interventional Radiology. The technology is its early stages, but mayserve as an adjunct to existing methods of training. In addition to stereoscopic 3D tutorials, interactive 360-degree video contentmay prove valuable to those unfamiliar with an interventional suite.

ParticipantsJason M. Hostetter, MD, Baltimore, MD (Presenter) Nothing to DiscloseChristopher Trimble, MD, MBA, Baltimore, MD (Abstract Co-Author) Nothing to DiscloseMichael A. Morris, MD, MS, Baltimore, MD (Abstract Co-Author) Nothing to DiscloseJean Jeudy JR, MD, Baltimore, MD (Abstract Co-Author) Nothing to Disclose

Background

Radiology education has long consisted of static images displayed alongside text, often in powerpoint slides or textbooks. This

SSA11-04 Simultaneous Human-Readable and Structured Data Encoding from PDF Forms


SSA11-05 Call Cases Dashboard: What a First-Year Radiology Resident Knows before Call


presentation style is incongruent with the modern radiologist's digital, interactive workflow. Our project aims to allow radiologists toeasily create interactive, anonymized, standards compliant teaching files, available anywhere, on any device.

Evaluation

Using standard web technologies of HTML5 and JavaScript, we created a cloud-based application to create and store radiologycases. Integrated with our enterprise PACS via DICOM standards, studies are fully anonymized with one click, creating a web-basedteaching case that is always-on, shareable, and available to desktop PCs, tablet devices, and modern smartphones withoutinstalling third party software. Similar to normal web pages, cases can be manipulated with hyperlinks in case descriptions or inother web pages in which the case is displayed. Cases are viewed in a web browser with full DICOM support allowing a true PACSenvironment. The application has been used at our institution over the past year to create educational modules for medicalstudents as well as to create interactive educational e-books about radiology and pathology topics. Residents and faculty alsoutilize the system to maintain personal case logs and share interesting cases via email or text within HIPAA compliance.

Discussion

Our implementation brings full PACS functionality to a portable and extensible educational platform. By utilizing existing web andDICOM standards, any device with a modern web browser is a fully enabled DICOM viewer, and any standards compliant PACS canbe easily integrated to generate anonymized cases in a single click. A cloud based solution also allows easy sharing andcollaboration, by simply sending a link via email, text, or social media, in same way one may share an interesting article. Ourarchitecture eliminates the need for local archival storage, ensuring cases are always instantly available.

Conclusion

We created a cloud based radiology educational resource using standard web technologies, with full PACS functionality, andautomated anonymization to bring radiology educational materials to the modern web.

ParticipantsZachary S. Delproposto, MD, Detroit, MI (Presenter) Nothing to DiscloseMatthew C. Rheinboldt, MD, New Orleans, LA (Abstract Co-Author) Nothing to Disclose

Background

Clinical and research needs often demand the need to store metadata which is not integral to the EHR (e.g., RECIST data). WhilePDF files can be saved as DICOM images, structure is lost and maintained only by entry into a separate system which requiresmaintenance and dissociates captured information from the study data. We have created a system to extract data from PDF Formsand simultaneously embed the form itself as a human-readable DICOM image, and the structured form data as an embedded QRcode within a DICOM image.

Evaluation

Using PDF forms eliminates the need for custom local or web applications to edit structured data. Storing the PDF data as an imageallows for facile retrieval of human-readable information sans additional software. QR-encoded structured data is stored as DICOMimage data, not using private/non-standard fields, maintaining superb vendor neutrality and compatibility. QR-codes also permitstructured data capture directly into mobile applications. Leveraging existing PACS infrastructure obviates the need for an externaldatabase. Extraction of structured data and PDF form re-creation occurs as needed; the presence of image-embedded structureddata does not diminish the performance of the PACS or impede interpretation of diagnostic data.

Discussion

Built using open-source components and leveraging the existing PACS infrastructure, we find that our system allows easy, reliablemetadata entry and recall (Figure 1). Since forms are stored both as a PDF image and as QR-encoded structured data image,workstations without the need for structured metadata entry or recall function normally without workflow impact. Most users arefamiliar with PDF forms, which are in common use and routinely used at both public and private institutions. PDF forms can beconstructed to constrain input fields to appropriate types (e.g., numeric fields), enforcing data consistency. Ease of use is a keyfactor in this system. Additional data storage requirements are minimal, adding only a few images to each study.

Conclusion

We show a robust, efficient, cost-effective method to encode metadata from PDF forms simultaneously in a human-readable imageand machine readable structured image formats.

ParticipantsLinda Kelahan, MD, Washington, DC (Presenter) Nothing to DiscloseAllan Fong, BS,MS, Washington, DC (Abstract Co-Author) Nothing to DiscloseRaj Ratwani, Washington, DC (Abstract Co-Author) Nothing to DiscloseRoss W. Filice, MD, Washington, DC (Abstract Co-Author) Nothing to Disclose

Background

Current ACGME guidelines do not help first-year radiology residents prepare for the daunting task of taking call. Tracking exposureto high-acuity cases that are likely to be encountered while taking call can provide a framework for focused study and reflection.Furthermore, these cases can subsequently be utilized for educational or research purposes.

Evaluation

We focused on cases most likely to be encountered in a resident call setting. First, we limited evaluation to relevant procedures(i.e. CT abdomen/pelvis for appendicitis). We then applied natural language processing (NLP) techniques (specifically sentence

SSA11-06 Resident Performance Analytics Using Structured Attending Feedback and #Hashtag Sharing Features


SSA11-07 Electronic Education Exhibit 'Likes' at the 2014 RSNA Annual Meeting


parsing) to identify both positive and negative cases where the diagnosis should have been considered and excluded irrelevantcases. Each night we apply these algorithms to and update our resident dashboard to reflect our findings. Residents can view thisdata in a custom date range with access to specific accession numbers for further review.

Discussion

Current methods of assessing resident progress, such as ACGME case log minimums, do not elucidate the types of pathology theresident sees nor the decisions the resident should make. Our model attempts to both highlight cases that have important call-relevant pathology but also negative cases where that pathology should be considered (i.e. CT for right lower quadrant pain). Bymaking this data available to our residents, we believe they will better understand how prepared they are for call, will haveopportunity to further educate themselves if needed, and will hopefully improve their receiver operating characteristic (ROC) oncall, which would positively impact clinical care. Furthermore, we are developing a repository of important call cases that anyresident could review for educational or research purposes.

Conclusion

Our "call cases" dashboard has multiple advantages to traditional CPT-code driven analysis of radiology resident education. We canidentify positive high acuity call-relevant studies viewed prior to beginning call, but also negative cases where there was highclinical suspicion. Furthermore, this repository allows residents to review call-relevant cases for educational or research purposes.

ParticipantsPo-Hao Chen, MD, MBA, Philadelphia, PA (Presenter) Nothing to DiscloseYin J. Chen, MD, Philadelphia, PA (Abstract Co-Author) Nothing to DiscloseMary H. Scanlon, MD, FACR, Haverford, PA (Abstract Co-Author) Nothing to DiscloseTessa S. Cook, MD, PhD, Philadelphia, PA (Abstract Co-Author) Nothing to Disclose

Background

The qualitative evaluation of a trainee's a radiology interpretation is an important dimension of training that is difficult tomeaningfully analyze and quantify. We previously presented an open-source web-based platform that provided volume-relatedanalytics for residents. Working from the existing code, we added visualization of attending feedback on resident reports, analyticsof qualitative grading of on-call studies, and social media features allowing annotation and sharing of interesting cases.

Evaluation

Building upon our open-source software, we implemented new functionality for analyzing attending modifications on both daytimeand off-hour preliminary interpretations. At our institution, attending radiologists grade independently-interpreted, on-call examsusing a 5-point scale: 'Great Call,' 'Agree,' 'Addition,' 'Minor Change,' and 'Major Change.' After loggin in, the software analyzes thetrainee user's attending grades and offers birds' eye view of discrepancy rates, turnaround times, and volume of call studies. Thedata can be organized by modality and anatomy. For each study, the software offers a 'Show Changes' view to highlight attendingedits on preliminary reports. An administrators' view is available for residency and fellowship directors. The trainee andadministrators may annotate each examination using free text, or apply social media-like #hashtags for sharing. We trackedtrainees' usage of the software before and after new feature implementation.

Discussion

Analysis of usage patterns during the three months post-implementation (9/1/2014-11/30/2014) showed 484 distinct user logins permonth, compared to 142 per month in the preceding three months (5/1/2014-7/31/2014) when these new features were notavailable. Additionally, 86 unique tags have been created by users with 962 total applications as of 2/28/2015. Some uses forhashtags include teaching ('#RareDx'), follow-up ('#FollowUp'), and documenting interpretive errors ('#SatOfSearch').

Conclusion

Implementation of new qualitative analytic and sharing features have led to substantial increases in usage of our resident analyticsplatform, suggesting that these features fulfill previously unmet educational needs.

ParticipantsPaul M. Bunch, MD, Boston, MA (Presenter) Nothing to DiscloseJeremy R. Wortman, MD, Boston, MA (Abstract Co-Author) Nothing to DiscloseKatherine P. Andriole, PhD, Dedham, MA (Abstract Co-Author) Advisory Board, McKinsey & Company, Inc;

PURPOSE

For the second consecutive year, the 2014 RSNA DPS offered electronic education exhibit (EEE) viewers the opportunity to "like" anEEE. We sought to 1) evaluate any relationship between an EEE's popularity and its chance of winning an award or being selectedfor RadioGraphics, 2) evaluate any relationship between an EEE's recognition and its subsequent popularity, and 3) assess overallaudience "like" participation at the 2014 meeting as compared to 2013.

METHOD AND MATERIALS

The number of likes each EEE received was recorded from DPS on 1) Wednesday morning before award selections and RadioGraphicsinvitations had been announced and 2) Saturday morning after the meeting had concluded. Data analysis was performed by meansof one-way ANOVA.

RESULTS

At the 2014 RSNA meeting, there were 1793 EEEs, which received 11074 likes (Mean 6.2, Min 0 [n=124], Max 109 [n=1]). Awardswere given to 404 EEEs (22.5%), which received 3452 likes (31.2%, Mean 8.5, Min 0 [n=13], Max 109 [n=1]). RadioGraphics

SSA11-08 Structured Reporting of Focal Lesions in the Abdomen to Assess Radiology Trainees' PerformanceDemonstrates Decreased Detection Errors for Suspicious Lesions with Increased Training

Sunday, Nov. 29 11:55AM - 12:05PM Location: S403A

SSA11-09 TROVE: Open Source Resident Dashboard with Disease Classification Using Natural LanguageProcessing (NLP) and Machine Learning (ML)

Sunday, Nov. 29 12:05PM - 12:15PM Location: S403A

invitations were given to 169 EEEs (9.4%), which received 1525 likes (13.8%, Mean 9.0, Min 0 [n=2], Max 54 [n=1]). EEEs receivingawards had significantly more likes prior to award selection (Mean 4.4 vs 3.5, p=0.0035) and at the end of the meeting (Mean 8.5vs 5.5, p<0.0001) than non-awarded EEEs. EEEs receiving RadioGraphics invitations had significantly more likes prior to invitationannouncement (Mean 4.6 vs 3.6, p=0.0272) and at the end of the meeting (Mean 9.0 vs 5.9, p<0.0001) than non-invited EEEs.Recognized EEEs received significantly more likes over the second half of the meeting than non-recognized EEEs (Mean 4.2 vs 1.9for awarded vs non-awarded, Mean 4.5 vs 2.2 for invited vs. non-invited, p<0.0001 for both). There was a 152% increase in totalEEE likes recorded at the 2014 RSNA meeting as compared to 2013 (11074 vs 4391).

CONCLUSION

The DPS "like" feature at the 2014 RSNA meeting allowed for substantial audience feedback on EEEs, with over 11000 distinctentries made. There was an association between EEE likes and EEE recognition, and EEE recognition was also associated with asubsequent increase in EEE likes. As compared to 2013, EEE viewers at the 2014 meeting recorded 2.5 times more likes.

CLINICAL RELEVANCE/APPLICATION

EEE likes gauge radiologists' opinions of EEEs and may predict awards and RadioGraphics invitations. The DPS like feature was moreutilized at the 2014 RSNA meeting as compared to the previous year.

ParticipantsJoe C. Wildenberg, MD,PhD, Philadelphia, PA (Presenter) Nothing to DisclosePo-Hao Chen, MD, MBA, Philadelphia, PA (Abstract Co-Author) Nothing to DiscloseCharles E. Kahn JR, MD, MS, Philadelphia, PA (Abstract Co-Author) Nothing to DiscloseHanna M. Zafar, MD, Philadelphia, PA (Abstract Co-Author) Nothing to DiscloseTessa S. Cook, MD, PhD, Philadelphia, PA (Abstract Co-Author) Nothing to Disclose

PURPOSE

Structured reporting (SR) of focal masses in the solid abdominal organs can be used, in the context of education, to assess theability of of radiology trainees to detect and characterize these lesions. Using an existing SR initiative at our institution, weinvestigated if there was a difference in detection of focal abdominal mass lesions by trainee level.


All CT and US studies of the abdomen performed between 7/1/2013 and 12/15/2014 in a call setting, without immediate attendinginput, were reviewed. Trainees evaluated the liver, pancreas, kidneys, and adrenal glands within the SR framework. Numeric codeswere analogous to BI-RADS, and corresponded to both the presence of focal masses and the likelihood of malignancy. All preliminaryinterpretations were subsequently reviewed by attending radiologists, and differences in the numerical categories noted. Non-visualization was representative of a change from an assignment of "no mass" to any benign, indeterminate or suspicious lesion.Data was analyzed by level of training.

RESULTS

Among 12081 studies that met inclusion criteria, residents failed to visualize focal abdominal masses more often than fellows(80/3699, 2.2% versus 128/8382, 1.5%, respectively) (p<0.02). Sub-analysis revealed no difference in detection when the lesionwas classified as benign; however, fellows demonstrated a lower miss rate for suspicious lesions (p<0.05). Furthermore, althoughdirect year-to-year comparisons were not significant, there was a near-linear decrease in non-visualization rate with increasedtrainee year (r=-0.96; p<0.05).

CONCLUSION

SR can be leveraged to assess radiology trainees' performance and guide education in call situations. We found that increasedtraining was associated with a lower proportion of missed focal masses. Additionally, we found that inexperienced trainees mademore errors when the lesion was eventually classified as suspicious, whereas there was no difference for benign lesions.


Trainee education in the clinical setting is often subjective, with incomplete metrics to assess trainee performance. Structuredreporting of focal abdominal masses is an objective metric to understand the progression of trainees' proficiency and providetargeted feedback for studies read in a call setting. The effect of alternative methods of targeted educational outreach toresidents on non-visualized abdominal masses can be explored.

Honored Educators


Charles E. Kahn JR, MD, MS - 2012 Honored Educator

ParticipantsKurt T. Teichman, BSc, MEng, New York, NY (Abstract Co-Author) Nothing to DiscloseShlomo Minkowitz, BA, MD, New York, NY (Presenter) Nothing to DiscloseCharles Herrmann, MS, New York, NY (Abstract Co-Author) Nothing to DiscloseGeorge L. Shih, MD, MS, New York, NY (Abstract Co-Author) Consultant, Image Safely, Inc; Stockholder, Image Safely, Inc;

Consultant, Angular Health, Inc; Stockholder, Angular Health, Inc;

Background

Determining if a radiology resident has met the goals of the residency curriculum, which outlines different pathologies and conditionsfor imaging, is challenging. Our TROVE dashboard (DEMO: http://demo.trovedashboard.com / SOURCE:http://src.trovedashboard.com) displays the volume of exams and now classifies the radiology reports to determine which diseasesa resident has seen during clinical rotations.

Evaluation

Machine learning algorithms are applied using a training dataset from our billing company which provides ICD9 codes for all radiologyreports.The disease view of the TROVE dashboard utilizes 174 support vector machine (SVM) classifiers to associate radiology freetext reports with 174 specific disease labels, which correspond to diseases that were determined to be important for residenttraining by the different departmental division chiefs.Resident reports are processed using Natural Language Processing techniquesby finding both positive and negative labels for a particular disease and then compiling a MESH concept list of the most commonconcepts associated with that particular disease to determine features to be used in SVM training.Utilizing solely the impressiontext of a report, the average F1-score across all 174 svm classifiers is 0.801894 (best = 1.0). Some diseases scored higher such asCarotid Stenosis (0.883001144096) and Wrist Fracture (0.886353355114). Other diseases had lower scores such as TesticularTorsion (0.676691729323). A single report may include multiple diseases and will be classified as such by the SVMs.

Discussion

With an average F1-measure of 0.801894 we can conclude that the methodology outlined above provides fairly robust predictionsassociated with the SVM classifiers, giving a reasonable estimate of resident experience. Lower scores are generally associatedwith either improper feature selection and/or shortage of training data which may improve over time with more data. Future work onthese SVM classifiers should include the use of MetaMap for better feature selection.

Conclusion

TROVE dashboard, using NLP and ML, provides classification of radiology reports to determine the diseases residents have seenduring their clinical rotations.

SPCP21A Opening Remarks

SPCP21B Roentgen - An X-Ray Journey

SPCP21C What Is Normal? Reference Values Derived from Population-Imaging and Their Role in ClinicalPractice

SPCP21

Germany Presents: Population-based Imaging: How Broader Research Efforts Can Affect Everyday Care andPrevention

Monday, Nov. 30 10:30AM - 12:00PM Location: E353C

OT RS


FDA Discussions may include off-label uses.

ParticipantsNorbert Hosten, MD, Greifswald, Germany (Moderator) Institutional research agreement, Siemens AG; Institutional researchagreement, Bayer AG; Stockholder, Siemens AGGabriele A. Krombach, MD, Aachen, Germany (Moderator) Nothing to Disclose

LEARNING OBJECTIVES

1) Appraise the contribution which Population-based Imaging can make to radiological knowledge. 2) Differentiate betweenclassical, prospective double-blind studies and the epidemiological, non-interventional approach to generate radiological knowledge.3) Assess information regarding normal findings, normal range and the like as generated by Population-based Imaging studies.

ABSTRACT

The „SHIP" (Study of health in Pomerania, Germany) has allowed to do more than 2000 whole-body MR scans in normal subjects inthe setting of an on-going epidemiological study during several years. An body of knowledge regarding the organization ofPopulation MR Imaging, the handling of incidental findings, the range of normal imaging findings and of imaging-related biomarkershas been generated. The course presents information about normal contrast enhancement patterns in the breast generated in alarge group (> 500); about MR findings both pathological and non-pathological that may be made in individuals in the absence ofdisease; about the distribution of quantitative parameters in cardiac imaging (plain and enhanced) in subjects in the absence ofovert heart disease. The success of the SHIP has encouraged to perform a similar, nation-wide study in Germany on an even largerscale. 5 centers have started to perform whole-body MRI in study participants. A large body of information on health status of theparticipants is generated by epidemiologists. Follow-up will be performed on a regular base in the frame of the so-called „NationalCohort". Information on the value of radiological methods will be generated by epidemiological methods, namely long-time follow up.

URL

Sub-Events

ParticipantsRonald L. Arenson, MD, San Francisco, CA (Presenter) Nothing to DiscloseNorbert Hosten, MD, Greifswald, Germany (Presenter) Institutional research agreement, Siemens AG; Institutional researchagreement, Bayer AG; Stockholder, Siemens AG

LEARNING OBJECTIVES

1) Appraise the contribution which Population-based Imaging can make to radiological knowledge. 2) Recommend Population basedMR Imaging as a vauable part of Population based epidemiological studies.

ABSTRACT

"Population-based MR Imaging" was chosen as the topic for this year's RSNA "Germany presents:" session. In Fermany, whole-bodyMRI is performed both in a regional study (Study of Health in Pomerania - "SHIP") and in the "National Cohort" which just started.The session explains (1) how normal ranges for contrast enhancement can be established in very large numbers of healthysubjects; (2) what "incidental" (or in the case of MRI patients) "unexpected" findings may be found on whole body MRI, (3) howwhole-body MRI may be set up in epidemiological population-based studies.

Participants

LEARNING OBJECTIVES


ParticipantsKatrin Hegenscheid, MD, Greifswald, Germany (Presenter) Research Grant, Bayer AG; Research Grant, Siemens AG; Research Grant,XERA 3 Deutschland GmbH

LEARNING OBJECTIVES

1) How population-based data are used to establish reference values for clinical diagnostics. 2) Which methods and procedures arenecessary for standardized analysis of large amounts of image data. 3) Which reference values have been developed so far fromthe population-based Study of Health in Pomerania and what clinical significance they have.

SPCP21D Why Population-Imaging may Help in Advancing Radiology: The German National Cohort

SPCP21E No Need to Look for Incidental Findings? Role in Clinical and Research Settings

SPCP21F Biomarkers of Cardiac Function in Population-Based Studies

SPCP21G Discussion and Closing Remarks

ABSTRACT

Prospective, population-based studies investigate the interaction between genetic predisposition for a disease, exposure toenvironmental factors and disease risk. They are a prerequisite for the development of prevention strategies. In the last decadesdue to its non-ionizing, examiner-independent, and high-resolution nature MRI has been implemented increasingly in epidemiologicalresearch. In 2008, the Study of Health in Pomerania (SHIP) was the first prospective population-based cohort study that offered astandardized whole-body MRI protocol for 3,772 participants aged 21 to 90 years. The primary objective of epidemiologic whole-body MR imaging is to phenotype a large subset of participants and to establish a comprehensive morphologic and functionalimaging bio-repository. In this presentation we describe how this bio-repository is used to derive reference values from Population-Imaging and their role in clinical practice.Since manual segmentation of a three dimensional organ is a laborious, time-consuming,and examiner-dependent process, it was necessary to develop automated methods for 3D analysis of a large set of data andorgans, e.g. the lungs, the liver, and the breasts. Supported by these automated segmentation methods first studies on referencevalues were conducted. For example reference values for the ascending and descending aortic wall thickness were provided itsassociation with age was investigated. Reference values for the gray and white matter brain volume were provided and theinfluence of genes, exogenous noxae, or diseases were described. We not only describe how organ volumes but also tissue analyzesbased on population-based data are performed. Methods for MR based fat quantification of the liver and the pancreas weredeveloped and the prevalence of fatty organ degeneration and its causes was investigated in the normal population. In women theinfluence of anthropometric measures and menopausal status on the contrast enhancement of normal breast parenchyma wasinvestigated and how it influences image analysis. Finally, we will show how reference values for the anterior chest wall thicknessare used for the optimal design of protective devices and personal body armor and influence established trauma guidelines fordecompression of tension pneumothorax.

ParticipantsFabian Bamberg, MD, MPH, Munich, Germany, ([email protected]) (Presenter) Speakers Bureau, Bayer AG; SpeakersBureau, Siemens AG; Research Grant, Bayer AG; Research Grant, Siemens AG;

LEARNING OBJECTIVES


ParticipantsSabine Weckbach, MD, Heidelberg, Germany (Presenter) Nothing to Disclose

LEARNING OBJECTIVES

1) Will be able to define "Incidental Finding ("IF"). 2) Will understand the underlying ethical problematic of Ifs. 3) Will be aware ofpossible onsequences of IF. 4) Will know about differences between imaging modalities in detecting IFs. 5) Will understand and beable to describe the categorization of IFs. 6) Will be able to differentiate between Ifs in clinical setting and research environmentand understand the different roles of radiologist and researcher. 7) Will be shortly informed about different approaches to IFs indifferent population based studies. 8) Will be provided a summary of management recommendations of IFs in clinical and researchsetting.

ABSTRACT

All findings which arise in the context of radiological diagnostics, potentially affect the health of a subject and without intention todetection of the corresponding finding are considered as radiological incidental findings (IF). The prevalence of IFs is worldwideincreasing due to the wider usage of modern imaging techniques such as MRI and CT in routine clinical practice as well as due toinclude imaging such as whole-body MRI in large population-based cohorts. From medical perspective, there is a need to report IFsin cases of potentially clinically relevant findings that need further workup or therapy. However, it is generally known that IFs mayhave a direct influence on life of the affected patient/ participant. The reporting of radiological IF may lead to further (eveninvasive) diagnostics and treatment and cause severe anxiety of patients and study participants. Possibly, there might also resultinsurance and occupational issues from the reporting of IFs. Therefore, subjects must especially be protected from consequencesof false-positives findings. This highlights why a very responsible approach to the reporting of IFs is warranted. The management ofIFs in clinical routine is regulated by the guidelines of the different academic societies. The management of IFs in the setting ofresearch studies differs depending on various factors such as study design, health status of enrolled subjects, etc. So far, widedifferences in approaches to IFs in different population based studies are observed. The course will illustrate why in general IFsshould be disclosed to the imaged subject if they are potentially clinically relevant. It will demonstrate the differences between IFsin clinical setting and research environment and highlight the different roles of radiologist and researcher.

URL

ParticipantsMarc Dewey, MD, Berlin, Germany (Presenter) Research Grant, General Electric Company; Research Grant, Bracco Group; ResearchGrant, Guerbet SA; Research Grant, Toshiba Corporation; Research Grant, European Commission; Research Grant, German ResearchFoundation; Speakers Bureau, Toshiba Corporation; Speakers Bureau, Guerbet SA; Speakers Bureau, Bayer AG; Consultant, GuerbetSA; Author, Springer Science+Business Media Deutschland GmbH; Editor, Springer Science+Business Media Deutschland GmbH;Institutional research agreement, Siemens AG; Institutional research agreement, Koninklijke Philips NV; Institutional researchagreement, Toshiba Corporation; ; ; ; ; ; ; ; ;

LEARNING OBJECTIVES


ParticipantsGabriele A. Krombach, MD, Aachen, Germany (Presenter) Nothing to DiscloseJames P. Borgstede, MD, Colorado Springs, CO (Presenter) Nothing to Disclose

LEARNING OBJECTIVES


SSC05-01 Health Service, Policy and Research Keynote Speaker: How to Establish a New Imaging Modality withDecent Evidence in Clinical Practice: What the Radiology Community Can Learn from Cardiac CT

Monday, Nov. 30 10:30AM - 10:40AM Location: S102D

SSC05-02 Is There an Association between STARD Statement Adherence and Citation Rate?


SSC05-03 Carotid Imaging in Canada: A Cost-Effectiveness Analysis


SSC05

ISP: Health Service, Policy and Research (Evidence-based Medicine/Guidelines/Outcomes)

Monday, Nov. 30 10:30AM - 12:00PM Location: S102D

HP RS

AMA PRA Category 1 Credits ™: 1.50ARRT Category A+ Credit: 0


ParticipantsMarta E. Heilbrun, MD, Salt Lake City, UT (Moderator) Nothing to DisclosePaul P. Cronin, MD, MS, Ann Arbor, MI (Moderator) Nothing to Disclose

Sub-Events

ParticipantsFabian Bamberg, MD, MPH, Munich, Germany (Presenter) Speakers Bureau, Bayer AG; Speakers Bureau, Siemens AG; ResearchGrant, Bayer AG; Research Grant, Siemens AG;

ParticipantsMarc Dilauro, MD, MSc, Ottawa, ON (Presenter) Nothing to DiscloseMatthew D. McInnes, MD, FRCPC, Ottawa, ON (Abstract Co-Author) Nothing to DiscloseDaniel Korevaar, Netherlands, Netherlands (Abstract Co-Author) Nothing to DiscloseChristian B. Van Der Pol, MD, Ottawa, ON (Abstract Co-Author) Nothing to DiscloseJeffrey Quon, MD, Ottawa, ON (Abstract Co-Author) Nothing to DiscloseStefan Walther, Berlin, Germany (Abstract Co-Author) Nothing to DiscloseDarya Kurowecki, Ottawa, ON (Abstract Co-Author) Nothing to DiscloseWilliam Petrcich, MSc, Ottawa, ON (Abstract Co-Author) Nothing to DisclosePatrick M. Bossuyt, PhD, Amsterdam, Netherlands (Abstract Co-Author) Nothing to Disclose

PURPOSE

To determine if adherence to the STARD checklist is associated with post-publication citation rates.


A comprehensive search of multiple databases including PubMed, EMBASE and Cochrane was performed in order to identify publishedstudies that have evaluated adherence of diagnostic accuracy studies to the Standards for Reporting of Diagnostic Accuracy(STARD) statement. Each study was searched in PubMED and Reuters Web of Science to yield a date of publication, journal impactfactor (IF), and a citation rate (citations/month). Univariate correlations were performed to identify any association between postpublication citation rate and STARD score as well as impact factor. A multivariate analysis was performed to explore the effect ofjournal impact factor.

RESULTS

Our search included 1002 eligible articles from 8 studies. The median journal IF was 3.97 (IQR: 2.32-6.21), the median STARD scorewas 15 (IQR 12-18), and the median citation rate was 0.0073 citations/month (IQR 0.0032-0.017). A weak positive correlation ofSTARD score with citation rate was identified (r=0.096, p=0.0024). There is a moderate positive correlation between impact factorand citation rate (r=0.58, p<0.0001). A weak positive correlation of impact factor with STARD score was identified (r=0.13,p<0.0001). A multivariate analysis revealed that when the effect of impact factor is partialed out, the positive correlation ofcitation rate with STARD score does not persist (r=0.026, p=0.42).

CONCLUSION

There is a positive correlation between journal impact factor and citation rate as well as impact factor and STARD score. Whenadjusted for journal impact factor, the positive correlation of citation rate with STARD score does not persist.


The variation in journal citation rate is influenced primarily by journal impact factor and to a lesser degree by STARD score.

ParticipantsEli Lechtman, PhD, MSc, Toronto, ON (Presenter) Nothing to DiscloseAlan R. Moody, MD, Toronto, ON (Abstract Co-Author) Nothing to DiscloseKevin Chen, Toronto, ON (Abstract Co-Author) Nothing to DiscloseSylvia Urbanik, Toronto, ON (Abstract Co-Author) Nothing to DisclosePascal N. Tyrrell, PhD, Toronto, ON (Abstract Co-Author) Nothing to Disclose

SSC05-04 Computed Tomography and Magnetic Resonance Imaging of Peritoneal Metastases: SystematicReview and Meta-analysis


PURPOSE

Diagnosing carotid artery disease relies on accessible and cost effective imaging to provide an accurate measure of stenosis.Currently, doppler ultrasound (DUS) is considered the first line modality of choice for suspected stenosis, while MR angiography(MRA) is often used to confirm the diagnosis and plan surgical interventions. In this simulation study, we explored the costeffectiveness of MRA alone vs DUS followed by MRA, for diagnosing suspected stenosis.


Cost effectiveness analysis (CEA) was conducted using TreeAge Pro. Decision trees were modeled for three populations: those withstenosis less than 50%, those with stenosis between 50-69%, and those with stenosis above 70%. Based on the imaging findings,the decision trees included surgical intervention, medical management, or standard care arms. Effectiveness was measured in termsof quality adjusted life years accounting for surgery and complications, stroke, and medical management. Values for the relevantinput variables were extracted from the literature, except the cost of imaging, which was reported from our institution.

RESULTS

Based on the CEA, MRA as a first line modality was more cost effective in populations with a high pretest probability of severestenosis >70%. In a clinical setting, this would reflect patients with multiple risk factors for carotid disease, or patients presentingwith symptoms of carotid stenosis such as a transient ischemic attack (TIA). While DUS as a first line modality was more costeffective for imaging the majority of patients suspected of having carotid stenosis <70%, CEA sensitivity analysis indicated thatreducing MRA costs by shortening MRA protocol time and increasing effectiveness of information reported, MRA as a first linemodality could be cost effective for an even larger portion of the at-risk population.

CONCLUSION

MRA alone may be more cost effective for patients with a high pretest probability of severe stenosis. Future simulations will explorethe effect of wait times on cost effectiveness, as well as the cost effectiveness of emerging MR imaging techniques to identifyplaque characteristics for stroke risk stratification and treatment decision making.


Magnetic resonance angiography is shown to be a cost effective first line imaging modality to assess carotid disease, providedthere is a high pretest probability of finding severe carotid stenosis.

AwardsTrainee Research Prize - Resident

ParticipantsDavide Bellini, MD, Latina, Italy (Presenter) Nothing to DiscloseDamiano Caruso, MD, Rome, Italy (Abstract Co-Author) Nothing to DiscloseMarco Rengo, MD, Rome, Italy (Abstract Co-Author) Nothing to DiscloseDomenico De Santis, MD, Rome, Italy (Abstract Co-Author) Nothing to DiscloseAndrea Laghi, MD, Rome, Italy (Abstract Co-Author) Speaker, Bracco Group Speaker, Bayer AG Speaker, General Electric CompanySpeaker, Koninklijke Philips NV

PURPOSE

Primary end point was to assess diagnostic accuracy of CT and MR in detecting Peritoneal Metastases (PM). Secondary end-pointswere determining sensitivity and specificity of CT scans in detecting PM for the thirteen regions according to Sugarbaker'sPeritoneal Cancer Index (PCI), investigating correlation between radiological PCI and surgical PCI, and comparing diagnostic yield ofCT versus PET/CT.


In June 2014, the MEDLINE, EMBASE, Cochrane Library, Sumsearch2 and Web of Science databases were searched. Methods foranalysis were based on PRISMA. Characteristics of patients and studies included were collected. QUADAS2 tool was used to assessthe methodological quality of the primary studies. Pooled estimates of sensitivity, specificity, positive and negative likelihood ratioswere calculated using fixed and random effects models. I2 was used to evaluate heterogeneity.

RESULTS

Twenty-two articles out of the 529 initially identified were selected (934 patients). Cumulative data of CT diagnostic accuracy onper patient basis were: Se 83% (95%CI: 79-86%; I2: 83.3%), Sp 86% (95%CI: 82-89%; I2: 65.5%), pooled positive LR 4.37 (2.58to 7.41; I2: 81.2%), pooled negative LR 0.20 (0.11 to 0.35; I2: 85.4%). On per region basis according to PCI, sensitivity of CT washigher in two regions: epigastrium, 78%% (95%CI 64-92%) and pelvis, Se 74% (95%CI 64-83%). Correlation between CT-PCI scoreand Surgical-PCI score were high, ranging from 0.49 to 0.96. MRI and PET/CT showed similar diagnostic accuracy of CT on perpatient basis.

CONCLUSION

By a good overall diagnostic accuracy on per patients basis and on per region basis according to PCI, CT should be considered theimaging modality of choice in patients affected by PM.


The role of imaging in detection of peritoneal metastases (PM) is still under debate. A systematically evaluation of diagnostic yieldof imaging modality is required to provide a better evidence-based advice to physicians in this area. CT should be considered theimaging modality of choice in patients affected by PM. Because of the good overall diagnostic accuracy on per region basisaccording to PCI, CT may lead surgeons to refer the patient to the best treatment option. MRI and PET/CT, at the moment, shouldbe considered second choices and further investigations are recommended.

SSC05-05 Computed Tomography (CT) in the Emergency Department: A Real-Time Study of Changes inPhysician Decision-Making


SSC05-06 Quenching MRI Anxiety: Complementary Alternative Medicine for Magnetic Resonance ImagingAnxiety


ParticipantsPari Pandharipande, MD, MPH, Boston, MA (Presenter) Nothing to DiscloseAndrew T. Reisner, MD, Boston, MA (Abstract Co-Author) Nothing to DiscloseWilliam D. Binder, MD, Providence, RI (Abstract Co-Author) Nothing to DiscloseAtif Zaheer, MD, Baltimore, MD (Abstract Co-Author) Nothing to DiscloseMartin L. Gunn, MBChB, Seattle, WA (Abstract Co-Author) Research support, Koninklijke Philips NV; Spouse, Consultant, WoltersKluwer NV; Medical Advisor, TransformativeMed, Inc; Ken F. Linnau, MD, MS, Seattle, WA (Abstract Co-Author) Speaker, Siemens AG; Royalties, Cambridge University Press; Chad M. Miller, MD, Durham, NC (Abstract Co-Author) Nothing to DiscloseLaura L. Avery, MD, Boston, MA (Abstract Co-Author) Nothing to DiscloseMaurice S. Herring, San Francisco, CA (Abstract Co-Author) Nothing to DiscloseAngela C. Tramontano, MPH, Boston, MA (Abstract Co-Author) Nothing to DiscloseEmily C. Dowling, Boston, MA (Abstract Co-Author) Nothing to DiscloseHani H. Abujudeh, MD, MBA, Boston, MA (Abstract Co-Author) Nothing to DiscloseJonathan D. Eisenberg, BA, Boston, MA (Abstract Co-Author) Nothing to DiscloseElkan F. Halpern, PhD, Boston, MA (Abstract Co-Author) Research Consultant, Hologic, IncKaren Donelan, DSc, Boston, MA (Abstract Co-Author) Nothing to DiscloseG. Scott Gazelle, MD, PhD, Boston, MA (Abstract Co-Author) Consultant, General Electric Company Consultant, Marval BiosciencesInc

PURPOSE

To determine how physicians' diagnoses, diagnostic uncertainty, and management decisions are affected by CT in emergencydepartment (ED) settings.


In this prospective, four-center study, ED patients referred to CT with abdominal pain, chest pain/dyspnea, or headache wereidentified. Before CT, physicians were surveyed to obtain their leading diagnosis, diagnostic confidence (0-100%), an alternative"rule out" diagnosis, and management plan (were CT not available). After CT, surveys were repeated. Primary measures includedproportions of patients for which leading diagnoses or admission decisions changed, and median changes in diagnostic confidence.Secondary measures addressed alternative diagnoses and return-to-care visits (e.g. to the ED) at one-month follow-up. Regressionanalysis identified associations between primary measures and site and participant (physician and patient) characteristics.

RESULTS

Paired surveys were completed for 1503 patients by 265 physicians. For abdominal pain, chest pain/dyspnea, and headache, leadingdiagnoses changed in 51% (278/545), 44% (208/471), and 25% (122/487) of patients. Pre-CT diagnostic confidence wasconsistently, inversely associated with the likelihood of a diagnostic change (p<0.0001). Median changes in confidence weresubstantial (+25%, +20%, +13% (p<0.0001)); median Post-CT confidence was high (95%, 93%, 95%) (Fig. 1). When reported, CThelped to confirm or exclude 'rule out' diagnoses in 95% or more of patients (96% (411/428), 97% (382/393), 95% (392/414)).Admission decisions changed in 25% (134/542), 18% (86/471), and 20% (94/480) of patients. During follow-up, 15% (82/545), 14%(64/471), and 10% (50/487) of patients returned for the same indication. Results correlated with site and participantcharacteristics in isolated circumstances.

CONCLUSION

Physicians' diagnoses and admission decisions changed frequently after CT, and valid diagnostic uncertainty was alleviated. Thesefindings suggest that current ordering practices are clinically justified.


For common referral indications to CT in emergency department settings, physicians' diagnoses and admission decisions changefrequently after CT, and valid diagnostic uncertainty is alleviated; these findings suggest that current ordering practices areclinically justified.

Honored Educators


Atif Zaheer, MD - 2012 Honored Educator

ParticipantsSelena I. Glenn, MA, BSRT, Portland, OR (Presenter) Nothing to Disclose

PURPOSE

Claustrophobia during MRI exams is a problem in imaging departments worldwide causing prematurely cancelled exams with financiallosses to medical facilities and delays patient care. A pilot study was conducted hypothesizing complementary alternative medicine(CAM) modalities aromatherapy and breathing techniques would decrease patient anxiety.


SSC05-07 Quest for More Personalized Lung Cancer Screening Strategy: Proximity of Smoking Cessation as aPredictor of Lung Cancer Events in High-risk Individuals Selected for Screening, Analysis withPropensity Score


SSC05-08 Quest for More Personalized Lung Cancer Screening Strategy: Age Older Than 54 Years as aPredictor of Lung Cancer Events in Individuals Selected for Screening


Thirty eight claustrophobic patients participated. They were four study arms, two experimental and two control groups.Experimental arms included participants who used anxiety medication (n=5), and non-medicated (n=13). The control arms includedparticipants who used anxiety medication (n=8) and, non-medicated (n=12).All scans except one were performed on a 1.5T wideshort bore scanner, and were of the hip region and above. Aromatherapy and breathing techniques was performed by theexperimental groups just before entering scanner bore. The control group was provided standard care and shamaromatherapy.Study theoretical schools of thoughts were integrative medicine and mixing humanistic and cognitive therapymethods. Study design was concurrent triangulation mixed methds. Quantitative data included Likert scales, physiological data andwere analyzed using an exact distributions based test, and regression analysis respectively. Qualitative data included open endedquestions analyzed by mapping common themes and quantified for histogram analysis.

RESULTS

A 76.5% statistically significant (p = .02 < 0.05) reduction in anxiety from pre scan anxiety to post CAM treatment in experimentalgroups, while control group experienced a statistically insignificant 66.7% (p = .12 >0.05) anxiety reduction. Likewise there was a76.5% (p = 0.02 < 0.05) average anxiety reduction in the experimental group during the MRI compared to pre scan levels, whilecontrol group anxiety reduction was not statistically significantly (p = 0.69 > 0.05). Qualitative data findings were 33% ofexperimental group said their anxiety was reduced, compared to 22% of the control group. Physiological data showed that as theheart rate increased the average anxiety increased.

CONCLUSION

Aromatherapy and breathing techniques may reduce anxiety during MRIs.


Fewer cancelled MRI exams with cost savings to medical facilities. Less interrupted medical treatment increasing patient carequality. A low cost skill based intervention for technologists.

ParticipantsRecai Aktay, MD, Pepper Pike, OH (Presenter) Nothing to DisclosePingfu Fu, Cleveland, OH (Abstract Co-Author) Nothing to DiscloseThomas Love, Cleveland, OH (Abstract Co-Author) Nothing to Disclose

PURPOSE

Purpose: To determine if proximity of smoking cessation (PoSC) is a predictor of incremental lung cancer events (LCE) among thosealready selected for lung cancer screening (LCS).


Methods and Materials: We stratified National Lung Screening Trial (NLST) cohort by PoSC (time from SC to randomization) intothree groups (>10 yrs and <5 identifying "remote-" and "recent-quitters" respectively). For each case, we estimated the propensity(PS) for remote-quitter using multivariable logistic regression (LR) -with 34 variables. From remote- (n=8,361) and recent-quitters(n=9,435), we produced 6,866 unique pairs of "remote-" and "recent-quitter" cases using PS matching. In the matched, and theentire groups of former smokers (FS) (n=27,692), we estimated the association between PoSC and incidences of LC and LC-death(LCD) using LR and restricted spline fit (RSF) of PoSC. We tested the models' goodness of fit (GOF) in quantiles of predictedprobabilities and calculated the area-under-the-curve (AUC) in ROC analysis for predictive performance.

RESULTS

Results: In the FS group, there were 149:331 respective LCD:LC cases of recent- and 98:205 cases of remote-quitters comparedto 102:244 and 69:145 LCD:LC cases respectively in the matched group. Recent-quitters were 71% more likely to have LC(OR=1.71;95%CI=1.39-2.12) and 50% more LCD (OR=1.50;95%CI=1.10-2.06) in the follow-up. Each proximate yr of SC isassociated with 4.8% increased risk for LC (OR=1.048;95%CI=1.032-1.065) and 4.5% for LCD (OR=1.045;95%CI=1.021-1.070). OnRSF, PoSC had significant (P<0.001 for LC and LCD), and linear associations with LC (P=0.788) and LCD (P=0.086). Validated andcalibrated LR models predicted LC and LCD with AUCs of 0.64 and 0.66 respectively with favorable GOF (P=0.739 for LC and 0.095for LCD).

CONCLUSION

Conclusion: In those already selected for LCS, the proximity of SC is linearly associated with increased risk for LCEs. Time-to-eventanalyses would explore the clinical usefulness of these relationships.


Clinical Relevance: A personalized LCS strategy may be devised through a second-round of risk profiling of those selected for LCSand PoSC may be used as one of the risk predictors in this endeavor.

ParticipantsRecai Aktay, MD, Pepper Pike, OH (Presenter) Nothing to DisclosePingfu Fu, Cleveland, OH (Abstract Co-Author) Nothing to DiscloseThomas Love, Cleveland, OH (Abstract Co-Author) Nothing to Disclose

PURPOSE

SSC05-09 Health Service, Policy and Research Keynote Speaker: Cost-effective Analyses

Monday, Nov. 30 11:50AM - 12:00PM Location: S102D

Purpose: To determine if age can increment the prediction of lung cancer events (LCE) in individuals who are already selected forlung cancer screening (LCS).


Methods and Materials: We stratified the National Lung Screening Trial cohort by age into three groups (>=64 years and 54-59identifying "senior-" and "young-group" respectively). For each case, we estimated the propensity (PS) for senior-group usingmultivariable regression (LR) -with 34 variables such as socio-demographic, exposure history,... From senior- (n=16,958) andyoung-groups (n=18,844), we produced 12,034 unique pairs of "senior" and "young" cases using PS matching. In the matched, andthe entire cohort (n=53,452), we estimated the association between participants' age and incidences of LC and LC-death (LCD)using LR and restricted spline fit (RSF) of age. We tested the models' goodness of fit (GOF) in quantiles of predicted probabilitiesand calculated the area-under-the-curve (AUC) in ROC analysis for predictive performance.

RESULTS

Results: In the entire group, there were 519:1016 and 203:422 respective LCD:LC cases in the senior- and the young-grouprespectively and in the matched group, 356:712 cases were senior and 129:286 cases were young.Seniors were more likely -thanyoungs- to have LC (OR=2.58;95%CI=2.24-2.97) and LCD (OR=2.78;95%CI=2.27-3.42) in the follow-up. In the entire group, LRshowed 8.7% increased risk of LC (OR=1.087; 95%CI=1.077-1.096) per year of age, however, this relationship was non-linear(P=0.0237) on RSF. For LCD, the risk increment was 8.9% per year (OR=1.089;95%CI=1.076-1.103) and this was linear (P=0.842)and significant (P<0.001).Calibrated LR with RSF predicted LC and LCD with AUCs of 0.63 and 0.68 respectively. GOF test wasfavorable with P-value of 0.421 for LC and 0.760 for LCD.

CONCLUSION

Conclusion: In those selected for LCS, age is a predictor of incremental LCEs. However, further time-to-event analyses are neededto determine the method for its potential clinical use.


Clinical Relevance: In those already selected for LCS, a second-round of risk profiling may allow the LCS strategy to be personalizedand age may be used as one of the predictors of LCEs in this process.

ParticipantsMarta E. Heilbrun, MD, Salt Lake City, UT (Presenter) Nothing to Disclose

RCC24

Overview of RSNA's Teaching File Software (MIRC®)

Monday, Nov. 30 2:30PM - 4:00PM Location: S501ABC

ED IN


ParticipantsWilliam J. Weadock, MD, Ann Arbor, MI (Presenter) Owner, Weadock Software, LLCStacy D. O'Connor, MD, Boston, MA (Presenter) Nothing to DiscloseAndre M. Pereira, MD, Toronto, ON (Presenter) Nothing to Disclose

LEARNING OBJECTIVES

1) Learn the features of the RSNA's MIRC software for teaching files. 2) Learn how to download and install the software. 3) Learnto use the RSNA MIRC Wiki to obtain documentation on the software.

ABSTRACT

Background:MIRC (Medical Imaging Resource Center) or TFS (Teaching File System) is a component of RSNA's CTP (Clinical TrialsProcessor), a suite of tools developed by RSNA to optimize research in radiology mainly with emphasis on:workflow and security ofpatient information. It is offered free of charge by RSNA.Simply put, MIRC can be:used to build a radiology teaching file, be it for anindividual of for an institution with many simultaneous users.Development started in 2000 and the project has been kept alive alongthe years, funded by RSNA, with great support both from RSNA and from the community of users.Installation is very streamlinedand available for virtually all plataforms and operational systems. All files necessary for installation are available at the downloadsession of RSNA's own MIRC server (http://mirc.rsna.org).This course is aimed to cover the basics of installation and administrationof MIRC and also basic and advanced authoring tools.After finishing this course the attendee will be proficient in authoring anduploading cases, and also be familiar with the resources for installation and administration of MIRC.Course outline:The followingtopics will be covered:1) MIRC overview2) Options of hardware3) MIRC Installation4) MIRC administration: setting uplibraries.5):MIRC administration: adding users6) Authoring a case using the basic authoring tool.7) Authoring a case using theadvanced authoring tool.8) Advanced authoring tools: image annotation, quizzes, adding documents.9) Other user-level tools:conferences, migrating cases stored in local folders10) Sending cases to MIRC straight from the Dicom viewer.After the talk theattendees will be granted access to an educational MIRC server which will be open for a full month after the conference, topractice authoring and uploading of cases.

RCA34

Using RSNA Clinical Trial Processing (CTP) Software for De-identification and Data Sharing (Hands-on)

Tuesday, Dec. 1 2:30PM - 4:00PM Location: S401AB

IN RS


ParticipantsJustin Kirby, Bethesda, MD (Presenter) Stockholder, Myriad Genetics, IncBradley J. Erickson, MD, PhD, Rochester, MN (Presenter) Stockholder, Evidentia Health, Inc; Stockholder, OneMedNet Corporation;Stockholder, VoiceIt Technologies, LLCKirk E. Smith, BS, Saint Louis, MO (Presenter) Nothing to Disclose

LEARNING OBJECTIVES

1) Learn about CTP's capabilities and the unique challenges associated with de-identifying DICOM images 2) Learn how to install theCTP software 3) Learn how to use Pipelines to quickly configure CTP for data sharing and clinical trial use cases 4) Learn how tocustomize de-identification scripts for advanced use cases

ABSTRACT

The RSNA Clinical Trials Processor (CTP) is free software that enables researchers to share data for imaging clinical trials andresearch projects. CTP provides a secure end-to-end solution for efficiently de-identifying and moving images and related databetween clinical trial sites or research teams. CTP is designed to support industry-standard Digital Imaging and Communications inMedicine (DICOM) transport protocols, so it is easy to configure CTP to work with commercial PACS systems as well as researchdatabases such as DCM4CHEE, NBIA, MIDAS or XNAT. Built-in compliance with DICOM de-identification standards enables easy andeffective removal of protected health information while preserving key attributes necessary to maintain usability of the data. In thiscourse participants will be provided with an overview of CTP's functionality and the unique challenges associated with de-identifying DICOM images. They will then perform hands-on image processing of sample data based on common research and clinicaltrial scenarios.

URL

SSJ12-01 Health Service, Policy and Research Keynote Speaker: Assessing Individual Performance in Radiology

Tuesday, Dec. 1 3:00PM - 3:10PM Location: S102D

SSJ12-02 Framing Bias Effects on Retrospective Reviews of Radiological Reports


SSJ12-03 Performance Testing for Radiologists Interpreting Chest Radiographs


SSJ12

ISP: Health Service, Policy and Research (Quality)


HP SQ RS

AMA PRA Category 1 Credit ™: 1.00ARRT Category A+ Credit: 1.00


ParticipantsJonathan James, BMBS, Nottingham, United Kingdom (Moderator) Nothing to DiscloseEdward Y. Lee, MD, MPH, Boston, MA (Moderator) Nothing to Disclose

Sub-Events

ParticipantsJonathan James, BMBS, Nottingham, United Kingdom (Presenter) Nothing to Disclose

ParticipantsJeffrey D. Robinson, MD, MBA, Seattle, WA (Presenter) Consultant, HealthHelp, LLC; President, Clear Review, Inc; Daniel S. Hippe, MS, Seattle, WA (Abstract Co-Author) Research Grant, Koninklijke Philips NV; Research Grant, General ElectricCompany

PURPOSE

When reviewing difficult exams, radiologists often disagree on the severity of a potential error. In the legal setting, this is oftenattributed to retrospective or framing bias. This study examines the effect of framing bias on radiologists' perceptions whenevaluating potential errors.


This study was IRB approved. Eleven de-identified exams that had been subject of malpractice litigation and four uncontestedcontrol exams were divided into four review sets each containing three litigation (L) exams and one control (C) and theiraccompanying reports. Volunteers solicited from the ACR directory were randomly assigned to one of four groups (P,D,Q,N). Group Pwas told that they had been retained by a malpractice plaintiff's attorney; D that they had been retained by a defense attorney; Qthat a neighboring hospital requested an outside QA review and N was given no context. Subjects were also randomly assigned toone of the four review sets, and asked for each exam if the radiology report failed to meet the standard of care (failure). The ratesat which each group judged each type of exam to be a failure were compared using a multivariate, mixed-effect, logistic regressionmodel.

RESULTS

The study was completed by 102 radiologists, yielding 368 reviews (276 L, 92 C).Together, all four groups rated L exams as failuresin 57% of assessments, and C exams in 27% (p= 0.006).The difference in ratings between L and C exams was most pronounced inGroup P(62% vs. 26%, p=0.013) and Group N(66% vs. 18%, p=0.003). Within the subgroup of L exams,Group P was significantlymore likely to judge an exam a failure than the Group D(62% vs 48%, p= 0.032). The Q and N groups were not significantly differentthan the other groups.

CONCLUSION

Framing bias plays a significant role in retrospective review. Told that the exams they were reviewing were problematic, reviewersrated 27% of control exams below the standard of care. Simulated plaintiff's experts rated litigation exams below the standard ofcare significantly more frequently that simulated defense experts rated the same exams. These differences in performance highlightthe effect such bias plays in actual expert witness review.


Since framing bias can significantly affect reviewers' impressions, blinding a reviewer to the nature of the exam being reviewedshould increase the objectivity of the reviewer's judgment.

ParticipantsYan Chen, Loughborough, United Kingdom (Presenter) Nothing to DiscloseJonathan James, BMBS, Nottingham, United Kingdom (Abstract Co-Author) Nothing to DiscloseLeng Dong, Loughborough, United Kingdom (Abstract Co-Author) Nothing to DiscloseAlastair G. Gale, PhD, Loughborough, United Kingdom (Abstract Co-Author) Nothing to Disclose

PURPOSE

The aim was to develop a system to assess the image interpretation performance of radiologists in identifying signs of malignancy

SSJ12-04 Do Socioeconomic Disparities Exist in Radiology? Multivariate Analysis of Socioeconomic FactorsImpacting Access to Imaging Services


SSJ12-05 Prevalence of Unanticipated Events Associated with MRI Examinations: A Benchmark for MRI Quality,Safety, and Patient Experience

on chest radiographs.


A test set of 30 digital chest radiographs was chosen by an experienced radiologist consisting of 11 normal and 19 challengingabnormal cases. The abnormal cases all had biopsy proven pathology; the normal cases had at least 2 years of imaging follow up.14 radiologists with a range of experiences were recruited. Participants individually read the test set displayed on a standardreporting workstation, with their findings entered directly onto a laptop running specially designed reporting software. For eachcase they were given the relevant clinical information and were asked to mark any perceived abnormality and rate their level ofsuspicion on a 5-points scale (normal, benign, indeterminate, suspicious or malignant). On completion of the test, participants weregiven instant feedback and had the opportunity to review cases were there was disagreement with the expert opinion andpathology. The time taken for the participants to complete the test was recorded.Differences between the participants'performance were assessed using ROC analysis.

RESULTS

The experience of the participants in reporting chest radiographs ranged from 1 to 26 years (Mean=9 yrs, Mdn=5 yrs). Participants'performance (ROC score) varied significantly between 2 groups (6 post-fellowship consultants, and 8 radiology residents). Radiologyresidents' performance as measured by ROC score was significantly poorer compared to post-fellowship consultants (Mean-RS=0.76, Mean-PFC=0.93, p=.003). There was a positive correlation between image interpretation performance (ROCMean=0.85,SD=0.11) and years of reading experience (Mean=9, SD=8.58) , r=.573, p=<.05, n=14.There was a trend for radiology residents totake longer to complete the task (Mean=26.51s) compared to post-fellowship consultant radiologists (Mean=19.65s), but this didnot quite reach statistical significance (p=.07).

CONCLUSION

This pilot study demonstrates that it is possible to devise a method for performance testing the reporting of chest radiographs.


Chest radiographs are the first line imaging test for patients with chest symptoms suspicious of malignancy, this pilot studydemonstrates that it is possible to devise methods to test performance of the reporting radiologist.

ParticipantsOmid Khalilzadeh, MD, MPH, Boston, MA (Abstract Co-Author) Nothing to DiscloseAlvin Y. Yu, MD, Boston, MA (Abstract Co-Author) Nothing to DiscloseEmmanuel Carrodeguas, BS, Boston, MA (Abstract Co-Author) Nothing to DiscloseAnand M. Prabhakar, MD, Somerville, MA (Abstract Co-Author) Nothing to DiscloseSynho Do, PhD, Boston, MA (Abstract Co-Author) Nothing to DiscloseGarry Choy, MD, MS, Boston, MA (Abstract Co-Author) Nothing to DiscloseJames A. Brink, MD, Boston, MA (Abstract Co-Author) Nothing to DiscloseEfren J. Flores, MD, Boston, MA (Presenter) Nothing to Disclose

PURPOSE

Racial disparities are known to exist in medicine, but little has been studied in radiology. One way to examine this is to look atmissed radiology appointments or missed care opportunities (MCO) which result in delayed diagnoses and negatively impact patientcare. Moreover, MCO in radiology may be a symptom of missed appointments in other specialties. The reason for missingappointments is multifactorial, and socioeconomic factors may play an important role. In this study, we investigated thedemographic factors associated with radiology missed appointments.


Demographic data of 975,539 ordered radiologic imaging exams at our institution in the calendar year 2014 was collected. Thedataset included: ethnicity/race, primary language, insurance status, and reasons for cancellation of the appointment. Theassociation of different factors with radiology MCOs was evaluated. Multivariate logistic regression models were implemented toevaluate the independent relationship between radiology MCOs and various factors.

RESULTS

MCO was the most common reason for not completing a radiologic exam (41.5%). Overall, there was about 5% MCO (42,854) inradiology appointments during the calendar year 2014. A primary language other than English (OR: 1.2), Black ethnicity (OR: 1.8,relative to White) and Hispanic ethnicity (OR: 1.5, relative to White) were significantly associated with higher odds of MCO on aradiology appointment. Among different scan type, the odds of MCO was significantly higher for CT angiograms (OR: 2.8, P<0.001).These associations remained significant after multiple adjustments for potential confounding variables.

CONCLUSION

There was a high number (42,854) of radiology MCO in the past year at our institution. Non-English primary language and Hispanicethnicity significantly correlate with likelihood of missing a radiology appointment. Our results identify patients who are at risk forMCO and provide opportunities for intervention that will improve the patient's experience and address healthcare disparities.Possible interventions to bridge the gap include telephone reminders in the patient's native language, scheduling radiologyprocedures with radiologists that come from similar background, assistance in coordination of transportation, among others.


Socioeconomic disparities exist in radiology. Further research in this area is paramount to examine the impact to healthcare access.


SSJ12-06 Technologist-directed Radiograph Repeats: Frequency and Associations


AwardsTrainee Research Prize - Resident

ParticipantsGelareh Sadigh, MD, Atlanta, GA (Presenter) Nothing to DiscloseAmit M. Saindane, MD, Atlanta, GA (Abstract Co-Author) Nothing to DiscloseKimberly E. Applegate, MD, MS, Zionsville, IN (Abstract Co-Author) Nothing to Disclose

PURPOSE

To determine the prevalence of unanticipated events (UE) associated with MRI examinations in a multi-center academic radiologydepartment.


UE reported by MRI technologists for examinations performed between June 2013 and November 2014 on 17 scanners in auniversity- (UH) and community-affiliated (CH) hospitals of single health system were retrospectively reviewed. Events werecategorized into: (1) orders and scheduling (no/improper order, insurance problem, scheduled wrong study/location, schedulingscreening failure, improper preparation instruction/study protocol); (2) delays in scan (late patient arrival/transport,anesthesia/pathology procedure delays, delays in getting correct protocol or checking images); (3) foreign bodies (unanticipatedmetal/foreign body/pacemaker); (4) non-contrast related (NONCON) patient events (claustrophobia, patient discomfort, bodyhabitus, pregnancy, nausea, pain, motion, need for sedation/general anesthesia, inability to complete the exam, patientdissatisfaction, patient fall, code called for resuscitation); (5) contrast related (CON) patient events (reaction, extravasation, lackof IV access, patient refusal of contrast); (6) technical acquisition issues (fat saturation, breath-holding, contrast bolus timing,mechanical scanner failure). Each category was compared between scanners located in UH vs. CH, and scanners that are solelyused for outpatient services (OP) vs. those used for outpatients and inpatients (OP/IP).

RESULTS

34,587 MRI examinations were assessed (87% UH; 59% OP) with 5,760 (17%) UE; (1.9% of patients had more than one categoryevents). Rates of UE for each category were as follows: 1.8% orders and scheduling [0.06% patient arriving wrong day, and 0.03%patient call-back], 3.3% delays in scan, 0.5% foreign bodies, 10.4% NONCON events, 1.3% CON events, and 1.5% technical issues.Most frequent patient issues were motion, claustrophobia, and need for sedation. UH exams had higher reported rate of UE. OPexams had higher rates of orders and scheduling problems and delays in scans, while OP/IP exams had more patient related andtechnical issues (all P<0.05).

CONCLUSION

UE associated with MRI exams are common (17%), with the majority being patient related issues.


Unanticipated patient events are common. Awareness of the prevalence and types of unanticipated events by MRI staff providesopportunities for practice improvement.

ParticipantsJill E. Jacobs, MD, New York, NY (Abstract Co-Author) Nothing to DiscloseAndrew B. Rosenkrantz, MD, New York, NY (Presenter) Nothing to DiscloseJoseph J. Sanger, MD, New York, NY (Abstract Co-Author) Nothing to DiscloseMarc Parente, New York, NY (Abstract Co-Author) Nothing to DiscloseDanny C. Kim, MD, White Plains, NY (Abstract Co-Author) Nothing to DiscloseMichael P. Recht, MD, New York, NY (Abstract Co-Author) Nothing to Disclose

PURPOSE

The decision to repeat a suboptimal radiograph by the technologist at the time of acquisition, prior to radiologist review, is aninfrequently assessed but potentially significant source of excess patient radiation. We assessed the technologist-directedradiograph retake rate in our hospital network.


We created an analysis tool to track all technologist-directed radiograph rejections for 52 CR and DR imaging device networks in 9of our hospital-based imaging centers. The tool captured all acquired images and the reject reason in a reject log file (RLF). All RLFswere downloaded monthly to an encrypted USB flash drive, renamed in standardized convention, and uploaded to a protectednetwork share drive. Information Technology staff reviewed all RLFs to ensure completeness and validity. RLFs were then importedinto a Reject Analysis Database. Analysis was performed for a 6 month period (6/1/14-11/30/14). Retake rate by case (RRC) wasnumber of retaken exposures (NR) acquired as a percentage of the total number of cases (TC) performed where RRC=(NR/TC)*100. Retake rate by exposure (RRE) was number of retaken exposures (NR) acquired as a percentage of the total numberof expected exposures (EE) for all performed examinations where RRE= (NR/EE)*100. Data was stratified by date, site, imagingdevice, body part, and reject reason.

RESULTS

Overall technologist-directed RRC and RRE were 3.4% and 1.8%, respectively. Body part RRC and RRE, respectively were: chest(5.9%, 4.4%); abdomen (3.3%, 1.6%); joint (3.0%, 1.3%); spine (2.6%, 1.2%); skull (1.8%, 1.0%); skeletal survey (1.6%, 0.8%),and unspecified (5.0%, 3.5%). For hospital portable devices, RRC was 9.2% overall (12.5% abdomen; 8.8% chest) and RRE was9.2% overall (10.8% abdomen and 9.0% chest). The most common reason for repeat exposures was positioning error (2.3% overall)for both portable and non-portable examinations.

CONCLUSION

Rates of technologist-directed radiograph retake vary by body part and are higher for portable examinations.


Technologist education to identify and correct sources of imaging error is necessary to reduce retake rates and decrease excesspatient radiation.

research and statistical methodsarchive.rsna.org/2015/researchandstatisticalmethods.pdfssa11-01...

Documents